Time-varying regression coefficients in Cox PH models -----------------------------------------------------
^stbtcalc^ varlist [^,^ ^w^indow^(^#^)^ ^i^ncrement^(^#^)^ ^sm^ooth stc > ox_options ]
^stbtplot^ [varlist] [^, ci le^vel^(^#^)^ graph_options ]
^stbtcalc^ calculates local estimates of the regression coefficients in the Cox model defined by ^stset^ with predictors in varlist. The estimates are calculated within moving windows which bracket a fixed number of "deaths" (events).
^stbtplot^ plots the estimates and confidence intervals against the failure times. varlist is (a subset of) the predictors in ^stbtcalc^'s varlist.
Options (^stbtcalc^) -------
^window(^#^)^ sets the window size. The larger the value of #, the smoother the estimated b(t) curve is, but the less far it reaches into the tails of the distribution of failure times. The default value of # is 10 * (number of xvars) or 20, whichever is the larger.
^increment(^#^)^ sets the incremental number of events. The event window is advanced by # failures. If # is set greater than 1, fewer Cox models are fitted and the execution time diminishes, which may be desirable in datasets with a large number of events. The default value of # is ((number of deaths) - (window size))/50 or 1, whichever is the larger. Provided there are enough deaths, this choice guarantees at least 50 local estimates of b(t).
^smooth^ smooths the estimated b(t) curve and its standard error using a running line smoother (see help @running@).
stcox_options are any of ^cluster()^, ^hr^, ^robust^.
Options (^stbtplot^) -------
^ci^ causes a ^level()^% confidence band to be plotted around the windowed estimates of b(t). The pointwise confidence intervals are based on the standard error of each b(t).
^level(^#^)^ sets the width of a pointwise confidence band for the estimated b(t). Default: current value of system macro ^$S_level^ (usually 95%).
graph_options are any of Stata's ^graph, twoway^ options
^stbtcalc^ provides a simple method of finding local estimates of the regressio > n coefficient for each covariate. All failures which occur outside the moving time window circumscribed by the minimum and maximum failure times of a fixed number of events become censored. Because of the mode of calculation, no estimates are available in the initial and final phases of follow-up.
^stbtcalc^ saves the local estimates of b(t) to new variables called ^BT^xvar1, ^BT^xvar2 etc, corresponding to the xvars. Standard errors are saved in other new variables called ^BS^xvar1, ^BS^xvar2, ... .
If more than one xvar is to be plotted by ^stbtplot^, a panel of plots is produced.
The smoothing for b(t) and its standard error (see the ^smooth^ option) uses @running@ with resmoothing option ^repeat(2)^.
. ^stset survtime cens^ . ^stbtcalc x1 x2 x3 x4 x5^ . ^stbtplot x2^ . ^stbtplot x1 x2 x3 x4, ci xlabel ylabel^
Patrick Royston Imperial College School of Medicine, London FAX: (011)-44-181-383-8573
Peter Sasieni Imperial Cancer Research Fund, London FAX: (011)-44-171-269-3429
Also see --------
Manual: [R] st stcox On-line: help for @stcox@, @stgtcalc@